Echoes of BERT: Do Modern Language Models Rediscover the Classical NLP Pipeline?

Michael Li, Nishant Subramani

Keywords: interpretability, probing, learned representations

Abstract: Large transformer-based language models dominate modern NLP, yet our understanding of how they encode linguistic information relies primarily on studies of early models like BERT and GPT-2. Building on prior BERTology work, we analyze 25 models spanning classical architectures (BERT, DeBERTa, GPT-2) to modern large language models (Pythia, OLMo-2, Gemma-2, Qwen2.5, Llama-3.1), probing layer-by-layer representations across eight linguistic tasks in English. Consistent with earlier findings, we find that hierarchical organization persists in modern models: early layers capture syntax, middle layers handle semantics and entity-level information, and later layers encode discourse phenomena. However, larger models compress this entire hierarchy toward earlier layer positions, suggesting they build richer representations more quickly. We dive deeper, conducting an in-depth multilingual analysis of two linguistic properties - lemma identity and inflectional features - that help disentangle form from meaning. We find that lemma information concentrates linearly in early layers but becomes increasingly nonlinear deeper in the network, while inflectional information remains linearly accessible throughout all layers. Additional analyses of attention mechanisms, steering vectors, and pretraining checkpoints reveal where this information resides within layers, how it can be functionally manipulated, and how representations evolve during pretraining. Taken together, our findings suggest that, even with substantial advances in LLM technologies, transformer models learn to organize linguistic information in similar ways, regardless of model architecture, size, or training regime, indicating that these properties are important for next token prediction.

Primary Area: interpretability and explainable AI

Submission Number: 23044